AI Detection

BuildVision AI uses machine learning to automatically detect and measure building components from your plans.

How It Works

1. Upload your PDF or image
2. Run AI Takeoff from the toolbar
3. Review detected items in the Results tab
4. Accept or reject detections as needed
5. Fill gaps with manual tools

Detectable Items

BuildVision AI can detect various building components:

Architectural

MEP (Mechanical, Electrical, Plumbing)

Site

Running AI Detection

Step 1: Select Detection Types

Click AI Takeoff in the toolbar and choose what to detect:

• All - Detect everything (takes longer)
• Architectural - Doors, windows, walls only
• MEP - Electrical, plumbing, HVAC only
• Custom - Select specific items

Step 2: Wait for Processing

Processing typically takes 30-60 seconds depending on:

• Drawing complexity
• Number of items to detect
• Server load

A progress indicator shows detection status.

Step 3: Review Results

Detected items appear in the Results tab:

• Green boxes - High confidence detections
• Yellow boxes - Medium confidence (verify these)
• Red boxes - Low confidence (likely wrong)

Working with Detections

Accepting Detections

• Individual: Click a detection, then click Accept
• Bulk: Use the sidebar to accept all of a type
• By confidence: Accept all high-confidence items at once

Rejecting Detections

• Individual: Click a detection, then click Reject or use Eraser tool (8)
• Bulk: Select multiple with Shift+click, then delete
• Clear all: Reset and start fresh

Editing Detections

AI detections can be edited like manual measurements:

1. Select the detection with the Select tool (1)
2. Drag vertices to adjust shape
3. Click edges to add new vertices
4. Double-click vertices to remove them

Trade Filter

Focus on what matters for your work by filtering detections by trade.

Enabling Trade Filter

1. Click Filter in the Results panel
2. Select trades to show:
   • Electrical
   • Plumbing
   • HVAC
   • Architectural
   • Fire Protection
   • Custom trades

Why Filter?

• Electricians see only outlets, fixtures, panels
• Plumbers see only fixtures, pipes, drains
• GCs can toggle between trades during review

Combined Views

Select multiple trades to see overlapping work:

• Electrical + HVAC for coordination
• Plumbing + Fire Protection for riser alignment
• All trades for complete picture

Assembly Integration

Filtered detections connect to trade-specific assemblies:

• Accept an electrical outlet → links to electrical assembly
• Accept a door → links to door hardware assembly
• Assemblies include labor and materials for that trade

Self-Learning

BuildVision AI improves from your corrections. The more you use it, the smarter it gets.

How It Works

1. AI makes a detection
2. You correct it (adjust, reject, or add missing)
3. System learns from your correction
4. Future detections improve

What the AI Learns

Feedback Quality

Better corrections = better learning:

• Precise adjustments help more than rough ones
• Consistent labeling trains better than random
• Correcting false positives reduces future noise

Viewing Learning Impact

Track how AI improves:

1. Go to Settings → AI Detection
2. See Learning Stats:
   • Corrections submitted
   • Accuracy trend over time
   • Most improved detection types

Privacy Note

Your corrections improve detection for your company only. Training data is not shared between companies.

Tips for Better Detection

Improve Your Drawings

• Use high-resolution PDFs (300+ DPI)
• Ensure symbols are clear and consistent
• Remove markup and revision clouds before scanning
• Use vector PDFs when possible (not scanned images)

Calibrate First

Set the scale before running AI detection. This helps the AI understand the drawing context and improves accuracy.

Run Multiple Passes

For complex drawings:

1. Run architectural detection first
2. Review and clean up
3. Run MEP detection
4. Final manual pass for missed items

Accuracy Expectations

AI detection is a starting point, not a final answer:

Troubleshooting

No detections found

• Check that the drawing has relevant content
• Try a different detection type
• Verify the image quality is sufficient

Too many false positives

• Use confidence filtering to hide low-confidence items
• Switch to a more specific detection type
• Consider manual takeoff for complex drawings

Detections are mis-sized

• Verify your scale calibration
• Check if the drawing uses non-standard symbols
• Manually adjust vertex positions

AI + Manual Workflow

The most efficient workflow combines AI and manual tools:

1. AI First: Run detection for bulk items (doors, windows, fixtures)
2. Quick Review: Accept obvious correct detections
3. Manual Clean-up: Fix misdetections and add missed items
4. Final Check: Verify quantities against plan notes

This hybrid approach typically saves 60-80% of manual effort while maintaining accuracy.